10.7.1.1 Zero-sum Rows and/or Columns

Note that if a row and/or column of the matrix to be factored contains all zeros then equally the factored matrix row/column will also be all zeros (under both singly and doubly constrained).

However confusion may arise if a row/column contains a very small number of very small valued cells (e.g., 0.0001). This may arise as a combination of rounding errors or if one matrix is subtracted from another. The problem is that the row/column sum may be printed as 0.00 (rounding down) which is therefore difficult to distinguish from a “true” 0.00 and, in this case, the factored row/column will not contain all zeros.

In release 10.9, (a), row/column sums less than 0.01 are printed with extra decimal places (e.g., 0.000012 rather than 0.00) and, (b), warning messages are printed under Furness to warn of very small row/column sums.